AI model decodes the language cells use to communicate.
Researchers at MIT and the Whitehead Institute built a machine learning model called IRIS that reads which chemical signals a developing cell received, using patterns of gene activity as a guide.
A team led by MIT biologist Pulin Li and graduate student Nicholas Hutchins has developed a machine learning model that can work out which signals a developing cell has received. Li is an assistant professor of biology and a member of the Whitehead Institute for Biomedical Research. The open-access study appeared in Nature Methods on Sept. 8.
As an embryo grows, a cell's fate depends on which genes are switched on or off. Cells guide one another by sending chemical signals through multi-step routes called signaling pathways, which convert outside messages into changes in gene activity. Retracing those instructions has been hard, the researchers explain, because scientists long assumed each pathway acts differently in each cell type and would need to be mapped separately in every one.
The team found that each signaling pathway leaves a distinctive "fingerprint" in gene activity, and that this fingerprint stays consistent across cell types. Their model, IRIS, is a neural network that examines a cell's overall gene activity and estimates which pathways were likely active at specific points in development. Li compared the approach to transfer learning, in which a voice recognition system trained mainly in English can use that training to recognize other languages. This is why, she said, IRIS can work on cell types it has not encountered before.
Li and Hutchins trained IRIS on an experimental dataset covering thousands of human embryonic stem cells. The cells had been exposed to dozens of combinations of six major signaling pathways at multiple developmental stages. The researchers then tested the model on single cells from mouse embryos during gastrulation, when cells rapidly branch into different fates. According to the team, IRIS accurately predicted when and where pathways would activate in cells headed toward heart, gut, muscle and spinal cord tissue.
In a further test, the researchers used IRIS to look for signals needed to produce a cell type important for lung development. The model predicted that activating one specific pathway would encourage lung-specific development, and experiments in mouse embryos confirmed it.
The team says the approach could speed up stem cell engineering and improve organoids, which are miniature 3D models of organs. Better lung models could be used to study asthma, lung cancer and pulmonary fibrosis, and to test potential therapies. Hutchins said IRIS helps researchers "decode the language cells use to talk to each other" faster than experiments alone allow.
At a Glance
- The study was published open access in Nature Methods on Sept. 8, led by Pulin Li and Nicholas Hutchins.
- IRIS was trained on data from thousands of human embryonic stem cells exposed to dozens of combinations of six signaling pathways.
- The model was tested on single cells from mouse embryos during gastrulation, including future heart, gut, muscle and spinal cord cells.
- IRIS predicted which pathway would promote lung-specific development, and experiments in mouse embryos confirmed the prediction.
Sources
This brief was written by Brightside Report from the reporting linked above. Spot an error? Request a correction.